Give me credit notebook

library(magrittr)
library(dplyr)
library(ROCR)
library(e1071)
library(randomForest)

# clear variables
closeAllConnections()
rm(list=ls())

# get new variables
creditDF <- read.csv("training.csv", sep = ",")
totalCreditDF <- creditDF
#creditDF <- creditDF[1:10000,]
head(creditDF)

plot(creditDF$RevolvingUtilizationOfUnsecuredLines) 

plot(creditDF$age) 

plot(creditDF$NumberOfTime30.59DaysPastDueNotWorse) 

plot(creditDF$DebtRatio) 

plot(creditDF$MonthlyIncome) 

plot(creditDF$NumberOfOpenCreditLinesAndLoans) 

plot(creditDF$NumberOfTimes90DaysLate) 

plot(creditDF$NumberRealEstateLoansOrLines) 

plot(creditDF$NumberOfTime60.89DaysPastDueNotWorse) 

plot(creditDF$NumberOfDependents) 

summary(creditDF$RevolvingUtilizationOfUnsecuredLines) 

summary(creditDF$age) 

summary(creditDF$NumberOfTime30.59DaysPastDueNotWorse) 

summary(creditDF$DebtRatio) 

summary(creditDF$MonthlyIncome) 

summary(creditDF$NumberOfOpenCreditLinesAndLoans) 

summary(creditDF$NumberOfTimes90DaysLate) 

summary(creditDF$NumberRealEstateLoansOrLines) 

summary(creditDF$NumberOfTime60.89DaysPastDueNotWorse) 

summary(creditDF$NumberOfDependents) 

# Take a sample to work on
creditDF <- totalCreditDF
creditDF <- creditDF[1:10000,]

creditDF <- creditDF %>% mutate(
                                  #1
                                  age = log(age),
                                  #2
                                  revolving.utilization.of.unsecured.lines = RevolvingUtilizationOfUnsecuredLines + 1,
                                  revolving.utilization.of.unsecured.lines = log(revolving.utilization.of.unsecured.lines),
                                  #3
                                  number.of.time.30.59.days.past.due.not.worse = NumberOfTime30.59DaysPastDueNotWorse + 1,
                                  number.of.time.30.59.days.past.due.not.worse = log(number.of.time.30.59.days.past.due.not.worse),
                                  #4
                                  number.of.time.60.89.days.past.due.not.worse = NumberOfTime60.89DaysPastDueNotWorse + 1,
                                  number.of.time.60.89.days.past.due.not.worse = log(number.of.time.60.89.days.past.due.not.worse),
                                  #5
                                  number.of.times.90.days.late = NumberOfTimes90DaysLate + 1,
                                  number.of.times.90.days.late = log(number.of.times.90.days.late),
                                  #6
                                  debt.ratio = DebtRatio + 1,
                                  debt.ratio = log(debt.ratio),
                                  #7
                                  monthly.income = MonthlyIncome,
                                  monthly.income = ifelse(is.na(monthly.income), 0, monthly.income),
                                  monthly.income = monthly.income + 1,
                                  monthly.income = log(monthly.income),
                                  #8
                                  number.of.open.credit.lines.and.loans = NumberOfOpenCreditLinesAndLoans + 1,
                                  number.of.open.credit.lines.and.loans = log(number.of.open.credit.lines.and.loans),
                                  #9
                                  number.of.dependents = NumberOfDependents,
                                  number.of.dependents = ifelse(is.na(number.of.dependents), 0, number.of.dependents),
                                  number.of.dependents = log(number.of.dependents + 1),
                                  #10
                                  number.real.estate.loans.or.lines = NumberRealEstateLoansOrLines + 1,
                                  number.real.estate.loans.or.lines = log(number.real.estate.loans.or.lines),
                                  
                                  #compose attributes
                                  low.age           = ifelse(age < 21, 1, 0),
                                  not.eligible      = ifelse(age > 60, 1, 0),
                                  
                                  no.dependents                     = ifelse(number.of.dependents == 0, 1, 0),
                                  has.dependents                    = ifelse(number.of.dependents > 0, 1, 0),
                                  
                                  years.of.age.per.dependent        = age / (number.of.dependents + 1),
                                  
                                  no.income                         = ifelse(monthly.income == 0, 1, 0),
                                  income.per.person                 = monthly.income / (number.of.dependents + 1),
                                  income.per.age                    = monthly.income / (age + 1),
                                  
                                  debt                                = log(debt.ratio * monthly.income + 1),
                                  debt.over.income                    = debt / (monthly.income + 1),
                                  no.debt                             = ifelse(debt.ratio == 0, 1, 0),
                                  debt.with.no.income                 = ifelse(debt > 0 && monthly.income == 0, 1, 0),
                                  debt.higher.33                      = ifelse(debt.ratio > .33, 1, 0),
                                  debt.higher.5                       = ifelse(debt.ratio > .5, 1, 0),
                                  undefigned.debt                     = ifelse(is.na(debt.ratio), 1, 0),
                                  debt.per.person                     = debt / (number.of.dependents + 1),
                                  debt.per.open.credit.lines          = debt / (number.of.open.credit.lines.and.loans + 1),
                                  debt.per.number.of.realestate.loans = debt / (number.real.estate.loans.or.lines + 1),
                                  
                                  remaining.income         = monthly.income - debt,
                                  
                                  extreme.credit.use       = ifelse(revolving.utilization.of.unsecured.lines > .99, 1, 0),
                                  no.credit.use            = ifelse(revolving.utilization.of.unsecured.lines == 0, 1, 0),
                                  
                                  credit.card.loans               = number.of.open.credit.lines.and.loans - number.real.estate.loans.or.lines,
                                  credit.card.loans.per.person    = credit.card.loans / (number.of.dependents + 1),
                                  
                                  real.estate.loans.per.person           = number.real.estate.loans.or.lines / (number.of.dependents + 1),
                                  revolving.loans.over.real.estate.loans = credit.card.loans / (number.real.estate.loans.or.lines + 1),
                                  
                                  late.minor.over.income           = number.of.time.30.59.days.past.due.not.worse / (monthly.income + 1),
                                  late.minor.over.remaining.income = number.of.time.30.59.days.past.due.not.worse / (remaining.income + 1),
                                  late.minor.over.debt             = number.of.time.30.59.days.past.due.not.worse / (debt + 1),
                                  
                                  late.mid.over.income           = number.of.time.60.89.days.past.due.not.worse / (monthly.income + 1),
                                  late.mid.over.remaining.income = number.of.time.60.89.days.past.due.not.worse / (remaining.income + 1),
                                  late.mid.over.debt             = number.of.time.60.89.days.past.due.not.worse / (debt + 1),
                                  
                                  late.major.over.income           = number.of.times.90.days.late / (monthly.income + 1),
                                  late.major.over.remaining.income = number.of.times.90.days.late / (remaining.income + 1),
                                  late.major.over.debt             = number.of.times.90.days.late / (debt + 1)
                                  
) %>% select(-Id, 
             -RevolvingUtilizationOfUnsecuredLines, 
             -NumberOfTime30.59DaysPastDueNotWorse, 
             -DebtRatio, 
             -MonthlyIncome, 
             -NumberOfOpenCreditLinesAndLoans, 
             -NumberOfTimes90DaysLate, 
             -NumberRealEstateLoansOrLines, 
             -NumberOfTime60.89DaysPastDueNotWorse, 
             -NumberOfDependents)
head(creditDF)

plot(creditDF$age) 

plot(creditDF$monthly.income) 

plot(creditDF$debt.ratio) 

plot(creditDF$number.of.open.credit.lines.and.loans) 

plot(creditDF$number.real.estate.loans.or.lines) 

plot(creditDF$revolving.utilization.of.unsecured.lines) 

plot(creditDF$number.of.time.30.59.days.past.due.not.worse) 

plot(creditDF$number.of.time.60.89.days.past.due.not.worse) 

plot(creditDF$number.of.times.90.days.late) 

plot(creditDF$number.of.dependents) 

summary(creditDF$age) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  3.045   3.714   3.951   3.912   4.127   4.615 

summary(creditDF$monthly.income) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  0.000   7.378   8.375   6.754   8.907  12.247 

summary(creditDF$debt.ratio) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.1601  0.3127  1.5153  0.6137 12.0367 

summary(creditDF$number.of.open.credit.lines.and.loans) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  0.000   1.792   2.197   2.075   2.485   3.850 

summary(creditDF$number.real.estate.loans.or.lines) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.0000  0.6931  0.5716  1.0986  2.8904 

summary(creditDF$revolving.utilization.of.unsecured.lines) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
0.00000 0.03066 0.15300 0.26065 0.45219 9.14217 

summary(creditDF$number.of.time.30.59.days.past.due.not.worse) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.0000  0.0000  0.1468  0.0000  4.5951 

summary(creditDF$number.of.time.60.89.days.past.due.not.worse) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
0.00000 0.00000 0.00000 0.04669 0.00000 4.59512 

summary(creditDF$number.of.times.90.days.late) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
0.00000 0.00000 0.00000 0.05835 0.00000 4.59512 

summary(creditDF$number.of.dependents) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.0000  0.0000  0.3896  0.6931  3.0445 

# normalization
creditDF <- creditDF %>% mutate(
                                  #1
                                  age                                          = (age - min(age)) / (max(age) - min(age)) ,
                                  revolving.utilization.of.unsecured.lines     = (revolving.utilization.of.unsecured.lines - min(revolving.utilization.of.unsecured.lines)) / (max(revolving.utilization.of.unsecured.lines) - min(revolving.utilization.of.unsecured.lines)) ,
                                  number.of.time.30.59.days.past.due.not.worse = (number.of.time.30.59.days.past.due.not.worse - min(number.of.time.30.59.days.past.due.not.worse)) / (max(number.of.time.30.59.days.past.due.not.worse) - min(number.of.time.30.59.days.past.due.not.worse)),
                                  number.of.time.60.89.days.past.due.not.worse = (number.of.time.60.89.days.past.due.not.worse - min(number.of.time.60.89.days.past.due.not.worse)) / (max(number.of.time.60.89.days.past.due.not.worse) - min(number.of.time.60.89.days.past.due.not.worse)),
                                  number.of.times.90.days.late                 = (number.of.times.90.days.late - min(number.of.times.90.days.late)) / (max(number.of.times.90.days.late) - min(number.of.times.90.days.late)),
                                  debt.ratio                                   = (debt.ratio - min(debt.ratio)) / (max(debt.ratio) - min(debt.ratio)),
                                  monthly.income                               = (monthly.income - min(monthly.income)) / (max(monthly.income) - min(monthly.income)),
                                  number.of.open.credit.lines.and.loans        = (number.of.open.credit.lines.and.loans - min(number.of.open.credit.lines.and.loans)) / (max(number.of.open.credit.lines.and.loans) - min(number.of.open.credit.lines.and.loans)),
                                  number.of.dependents                         = (number.of.dependents - min(number.of.dependents)) / (max(number.of.dependents) - min(number.of.dependents)),
                                  number.real.estate.loans.or.lines            = (number.real.estate.loans.or.lines - min(number.real.estate.loans.or.lines)) / (max(number.real.estate.loans.or.lines) - min(number.real.estate.loans.or.lines)),
                                  
                                  #compose attributes
                                  years.of.age.per.dependent             = (years.of.age.per.dependent - min(years.of.age.per.dependent)) / (max(years.of.age.per.dependent) - min(years.of.age.per.dependent)),
                                  income.per.person                      = (income.per.person - min(income.per.person)) / (max(income.per.person) - min(income.per.person)),
                                  income.per.age                         = (income.per.age - min(income.per.age)) / (max(income.per.age) - min(income.per.age)),
                                  debt                                   = (debt - min(debt)) / (max(debt) - min(debt)),
                                  debt.over.income                       = (debt.over.income - min(debt.over.income)) / (max(debt.over.income) - min(debt.over.income)),
                                  debt.per.person                        = (debt.per.person - min(debt.per.person)) / (max(debt.per.person) - min(debt.per.person)),
                                  debt.per.open.credit.lines             = (debt.per.open.credit.lines - min(debt.per.open.credit.lines)) / (max(debt.per.open.credit.lines) - min(debt.per.open.credit.lines)),
                                  debt.per.number.of.realestate.loans    = (debt.per.number.of.realestate.loans - min(debt.per.number.of.realestate.loans)) / (max(debt.per.number.of.realestate.loans) - min(debt.per.number.of.realestate.loans)),
                                  remaining.income                       = (remaining.income - min(remaining.income)) / (max(remaining.income) - min(remaining.income)),
                                  credit.card.loans                      = (credit.card.loans - min(credit.card.loans)) / (max(credit.card.loans) - min(credit.card.loans)),
                                  credit.card.loans.per.person           = (credit.card.loans.per.person - min(credit.card.loans.per.person)) / (max(credit.card.loans.per.person) - min(credit.card.loans.per.person)),
                                  real.estate.loans.per.person           = (real.estate.loans.per.person - min(real.estate.loans.per.person)) / (max(real.estate.loans.per.person) - min(real.estate.loans.per.person)),
                                  revolving.loans.over.real.estate.loans = (revolving.loans.over.real.estate.loans - min(revolving.loans.over.real.estate.loans)) / (max(revolving.loans.over.real.estate.loans) - min(revolving.loans.over.real.estate.loans)),
                                  late.minor.over.income                 = (late.minor.over.income - min(late.minor.over.income)) / (max(late.minor.over.income) - min(late.minor.over.income)),
                                  late.minor.over.remaining.income       = (late.minor.over.remaining.income - min(late.minor.over.remaining.income)) / (max(late.minor.over.remaining.income) - min(late.minor.over.remaining.income)),
                                  late.minor.over.debt                   = (late.minor.over.debt - min(late.minor.over.debt)) / (max(late.minor.over.debt) - min(late.minor.over.debt)),
                                  late.mid.over.income                   = (late.mid.over.income - min(late.mid.over.income)) / (max(late.mid.over.income) - min(late.mid.over.income)),
                                  late.mid.over.remaining.income         = (late.mid.over.remaining.income - min(late.mid.over.remaining.income)) / (max(late.mid.over.remaining.income) - min(late.mid.over.remaining.income)),
                                  late.mid.over.debt                     = (late.mid.over.debt - min(late.mid.over.debt)) / (max(late.mid.over.debt) - min(late.mid.over.debt)),
                                  late.major.over.income                 = (late.major.over.income - min(late.major.over.income)) / (max(late.major.over.income) - min(late.major.over.income)),
                                  late.major.over.remaining.income       = (late.major.over.remaining.income - min(late.major.over.remaining.income)) / (max(late.major.over.remaining.income) - min(late.major.over.remaining.income)),
                                  late.major.over.debt                   = (late.major.over.debt - min(late.major.over.debt)) / (max(late.major.over.debt) - min(late.major.over.debt))
                                
)
head(creditDF)

plot(creditDF$age) 

plot(creditDF$monthly.income) 

plot(creditDF$debt.ratio) 

plot(creditDF$number.of.open.credit.lines.and.loans) 

plot(creditDF$number.real.estate.loans.or.lines) 

plot(creditDF$revolving.utilization.of.unsecured.lines) 

plot(creditDF$number.of.time.30.59.days.past.due.not.worse) 

plot(creditDF$number.of.time.60.89.days.past.due.not.worse) 

plot(creditDF$number.of.times.90.days.late) 

plot(creditDF$number.of.dependents) 

summary(creditDF$age) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.4260  0.5773  0.5522  0.6893  1.0000 

summary(creditDF$monthly.income) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.6025  0.6839  0.5515  0.7273  1.0000 

summary(creditDF$debt.ratio) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
0.00000 0.01330 0.02598 0.12589 0.05099 1.00000 

summary(creditDF$number.of.open.credit.lines.and.loans) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.4654  0.5707  0.5390  0.6454  1.0000 

summary(creditDF$number.real.estate.loans.or.lines) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.0000  0.2398  0.1978  0.3801  1.0000 

summary(creditDF$revolving.utilization.of.unsecured.lines) 

    Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
0.000000 0.003354 0.016736 0.028510 0.049462 1.000000 

summary(creditDF$number.of.time.30.59.days.past.due.not.worse) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
0.00000 0.00000 0.00000 0.03195 0.00000 1.00000 

summary(creditDF$number.of.time.60.89.days.past.due.not.worse) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
0.00000 0.00000 0.00000 0.01016 0.00000 1.00000 

summary(creditDF$number.of.times.90.days.late) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.0000  0.0000  0.0127  0.0000  1.0000 

summary(creditDF$number.of.dependents) 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.0000  0.0000  0.1280  0.2277  1.0000 

d <- dim(creditDF)
trainingDF <- creditDF[1:(d[1] * .8), ]
testDF     <- creditDF[(d[1] * .8 + 1):d[1],]
head(trainingDF)

head(testDF)

model.lr <- glm(SeriousDlqin2yrs ~.,  family=binomial(link='logit'),  data=trainingDF)
pred.lr <- predict(model.lr, newdata=testDF, type="response")

prediction from a rank-deficient fit may be misleading

pr.lr <- prediction(pred.lr, testDF$SeriousDlqin2yrs)
prf <- performance(pr.lr, measure = "tpr", x.measure = "fpr")
plot(prf)

auc <- performance(pr.lr, measure = "auc")
auc <- auc@y.values[[1]]
auc

[1] 0.8099722

model.svm <- svm(SeriousDlqin2yrs ~ ., data=trainingDF)

Variable(s) ‘low.age’ and ‘not.eligible’ and ‘debt.with.no.income’ and ‘undefigned.debt’ constant. Cannot scale data.

pred.svm <- predict(model.svm, testDF, type="response")
pr.svm <- prediction(pred.svm, testDF$SeriousDlqin2yrs)
prf.svm <- performance(pr.svm, measure = "tpr", x.measure = "fpr")
plot(prf.svm)

auc <- performance(pr.svm, measure = "auc")
auc <- auc@y.values[[1]]
auc

[1] 0.6960223

auc

[1] 0.8099621

training.results.DF <- data.frame(lr = predict(model.lr, trainingDF, type="response"),
                                  #svm = predict(model.svm, trainingDF, type="response"),
                                  rf = predict(model.rf, trainingDF, type="response"),
                                  SeriousDlqin2yrs = trainingDF$SeriousDlqin2yrs)

prediction from a rank-deficient fit may be misleading

model.lr.fin <- glm(SeriousDlqin2yrs ~.,  family=binomial(link='logit'),  data=training.results.DF)

glm.fit: fitted probabilities numerically 0 or 1 occurred

test.fin.df <- data.frame(lr = pred.lr, 
                          #svm = pred.svm, 
                          rf = pred.rf, 
                          SeriousDlqin2yrs = testDF$SeriousDlqin2yrs)
pred.lr.fin <- predict(model.lr.fin, newdata=test.fin.df, type="response")
pr.lr.fin <- prediction(pred.lr.fin, test.fin.df$SeriousDlqin2yrs)
prf.fin <- performance(pr.lr.fin, measure = "tpr", x.measure = "fpr")
plot(prf.fin)

auc <- performance(pr.lr.fin, measure = "auc")
auc <- auc@y.values[[1]]
auc

[1] 0.8083532